Should you live near an MRT station? : A comparative study on Singapore private property market using hedonic and machine learning models

Abstract

Under Singapore Land Transport Authority Master Plan 2040, the agency has proposed the construction of three more Mass Rapid Transit (MRT) lines across Singapore. This study aims to evaluate how public transport networks have been capitalized into Singapore private housing market as premiums and how individuals should estimate costs and benefits when considering living closer to MRT stations. Using transaction data of all private property transactions with added features detailing distances to amenities and schools across 1995-2018, our research attempts to quantify the MRT distance premium with hedonic models consisting of 3 fixed-effects models on 4 different heterogenous subsample groups. In the meantime, an investigation using 5 machine learning models under 3 categories – LASSO, Random Forest and Artificial Neural Networks was conducted to address the same questions with deeper insights on importance of determinants of property prices. The results suggest that the MRT distance premium is significant and moving 100 meters closer from the mean distance point (603.61 meters) to the nearest MRT station will cause an increase of 15,131 SGD in the overall transacted price. Machine learning models generally achieved a higher prediction accuracy, and the interaction term with the property age was suggested by LASSO to improve the coefficient of determination. From results derived in Random Forest models, property prices are mostly affected by the broader macroeconomic factors during the time of sale, as well as the size and floor level of the property. Other important factors includes the ease of access to public transportation, living amenities around the property and the age of the property with distance to MRT station being the most important of these factors. An appraisal on different approaches was provided in the end as future implications for researchers to utilize additional data sources and data-driven models to exploit potential causal effects in economic studies.